I know it seems like because you have AI at your disposal, you can just jump right in and have it start producing code for your next big coding project. But trust me, from experience, there's a better way that pays off.
### 1. Plan
FIRST, use your AI platform's deepest thinking model to create your plan document with all the requirements. You feed it everything you want the application to do, including best practices and the like. Have it generate that document. This will take about 10 minutes of your time to come up with your reqirements, then the AI will spend maybe 90 seconds more, then you'll have a very well thought out and documented plan for your project.
### 2. Documents
Now, turn off the deep thinking (because it's expensive and slow) and go back to your AI platform's regular option that's best for coding. Add the document to your source project's docs folder and to your AI's project's knowledge repository (Claude & ChatGPT each support this, for the paid plans).
### 3. Code
Now it's time to start coding. Tell the AI to use that document throughout and to NOT veer from it for any reason. Treat the document as its coding Bible.
It's also a good idea to have it create a TODO.md file with checkboxes in it to keep track of what it needs to do and what it's completed. AI benefits from TODO lists too.
This will keep you AI on target and following the plan from start to finish.
No AI was used in writing this article. I mean, I tried, but Gemini was only producing the typical AI slop. So, I went old school on this one. You're welcome! And I'm sorry!